||Intel's Dual-Core Pentium Extreme Edition 840 Performance Pre-Preview
April 04, 2005 Chris Crazipper Angelini
Summary: 2005 will no doubt go down as the year for dual-core. With clock speeds beginning to taper off, both AMD and Intel are shifting in a different direction, integrating multiple cores on the same processor die. Intel was first to get a dual-core chip in our hands, the Pentium Extreme Edition 840. Read all about its design and new chipset platform as well as how it performns in comparison to previous Intel processors, as well as the competition in this article!
| Introduction||Page:: ( 1 / 14 )|
Enter dual-core technology. Both Intel and AMD believe that PC usage models are evolving towards more intensive multi-tasking. Multimedia, language processing, streaming content--theyíre all applications that will operate at the same time, consuming precious processing resources. Instead of exclusively devoting energy to ramping clock speed, an exercise of finite potential given current manufacturing techniques and power consumption issues, the two competitors decided within short succession that theyíd instead broaden their capacity for operating on multiple threads simultaneously.
Of course, both companies already sell intricate multi-processing platforms for the server and workstation markets. However, because they consist of multiple physical processors, expensive motherboards capable of accommodating extra sockets, and premium memory, those generally arenít viable options in a home or small business. Theyíd be of no use to the casual gamer, anyway. Existing titles are single-threaded and writing games to multiple threads promises to be a laborious process. Thatís why current Intel Xeon and AMD Opteron systems sell to higher-end markets.
The goal with dual-core is two-fold. First, enable the performance of two physical processors on one chip, in an evolution of what Hyper-Threading offered. Second, bring the cost of multi-processing down to a mainstream level--hopefully, equivalent to an existing single-core platform. Intel and AMD are going about their respective dual-core plans in slightly different ways, using divergent architectures and launching to different audiences.
AMD is of the opinion that the server and workstation markets stand to benefit from dual-core technology most immediately and as such are currently working on a dual-core Opteron processor for an early second-quarter launch. Intel, on the other hand, is gunning for the home users first with a dual-core Pentium 4 processor due out in the second quarter as well. Given todayís software landscape, it indeed appears that AMD stands to generate the most excitement with dual-core at the high-end, but thatís not why weíre here today. In a bid to prove its dual-core readiness, Intel recently shipped us its forthcoming Pentium Extreme Edition 840, a 3.2GHz running with an 800MHz FSB and a motherboard centering on its 955X Express chipset for a preview of whatís to come.
| Examining Intelís Dual-Core||Page:: ( 2 / 14 )|
Laying the Foundation
If anything, Intel is adamant that this equipment is production-ready. At the same time, Intel is saying that the 3.2GHz dual-core components do not represent the best single-threaded performance you can get. Existing Pentium 4 Extreme Edition processors--even the higher-clocked standard Pentium 4 chips--will deliver better performance in games and other single-threaded environments. But the general belief is that dual-core processors will pave the way for more multi-threaded software, and as a trend the chips will outperform existing architectures, even at lower clock speeds.
Naturally, thatís a significant disclosure for gamers because we know upfront that dual-core wonít necessarily make a positive impact on performance just yet. I had the chance to sit down with two gentlemen from Intelís benchmarking lab during CES earlier this year and we discussed the ramifications of dual-core in gaming. They were able to explain that 3DMark05ís CPU module favors Intelís design so heavily because it is multi-threaded, and benefits from Hyper-Threading. Dual-core is expected to amplify that advantage.
It also came out that while there currently arenít any multi-threaded games, content is in development. Writing and optimizing a game for multiple threads is significantly more difficult, it turns out. And although we may not see wide-spread use of parallel processing architectures until late 2006, physics, animation, sound, and AI are all areas where dual-core will imminently change the way games are played.
So why bother with dual-core now? Apparently, thereís more to life than games. Intel is pushing threaded performance harder than ever and dual-core technology promises better performance than Hyper-Threading. If you still doubt, consider that Intelís roadmap is currently loaded with multi-core designs, ranging from the 65nm mobile Yonah to the 1.7 billion-transistor Montecito core, intended for enterprise-level servers.
Given their respective processor architectures, Intel and AMD are going about dual-core in distinctly different ways. Weíll save the Opteron banter for a separate story except to say that AMD is comfortable enough with its current platform to enable dual-core processors as drop-in upgrades for existing motherboards. According to the company, its architecture was originally designed for the move to dual-core, which is only now possible, thanks to 90nm manufacturing.
Conversely, Intelís approach necessitates a brand new chipset and motherboard. Past conversations with Intel employees have suggested that the accompanying platform would better cope with dual-core throughput requirements. At least weíre seeing Intel stick with its LGA775 socket interface.
| More Intel Dual-Core||Page:: ( 3 / 14 )|
Apples to Two Apples
Each core on the die has its own execution resource pool, its own 1MB L2 cache repository, and its own bus interface. Unfortunately, as with Intelís Xeon family, both cores are forced to share available front side bus bandwidth and at least for now, dual-core chips will top out on 800 MHz buses. As a result, the limitations of Intelís current bus technology versus competing designs remain intact.
Many of the technologies recently exposed on the Pentium 4 are also part of the Pentium D. EM64T, for example, is seemingly now a staple in Intelís desktop arsenal. The Execute Disable bit, compatible with Windows XP Service Pack 2, is also present and accounted for.
|<% print_image("02"); %>||<% print_image("03"); %>|
Moving forward, Intel plans to leverage manufacturing advances with a 65nm dual-core processor codenamed Presler. Rather than employ one die with two cores, Presler leverage two cores on two separate dies co-existing on a single package. Splitting the two die apart shouldnít make a perceptible performance difference, but we can only assume it simplifies manufacturing somewhat, as the complexity of a single die is halved, improving yields. Presler isnít expected until the first half of 2006, but it, like Smithfield will continue communicating over a shared MCH bus.
Around the same time Intel says it will launch Paxville, the Xeon MP dual-core equivalent on a 90nm manufacturing process, and Dempsey, a 65nm Xeon DP dual-core chip at 65nm. The two server/workstation designs will interact with brand new chipsets, each sporting a pair of front side buses to counteract Intelís shared topology.
| Pentium Extreme Edition 840||Page:: ( 4 / 14 )|
Introducing a New Extreme Edition
Anxious to prove that itís ready and waiting with dual-core product, Intel sent a sample of its Pentium Extreme Edition 840, a Smithfield processor running at 3.2 GHz. Weíve mentioned in the past that Intel has had some trouble differentiating its Extreme Edition lineup from other Pentium processors. Even today, if you pay $1,000 for a 3.73 GHz, youíre merely treated to a faster front side bus. All of that should change with the Extreme Edition 800-series. Though it runs on the same 800 MHz bus, Extreme Edition processors now get exclusive rights to Intelís Hyper-Threading technology. Standard Pentium D processors, already laden with two cores, will be able to operate on two threads simultaneously, while the Extreme Edition leverages Hyper-Threading to power through four threads. Windows XP recognizes both physical cores in addition to the logical processors, displaying four threads in its task manager.
And while Intel isnít making a public spectacle of it, the Extreme Edition chips will also start shipping with unlocked clock multipliers, much like AMDís Athlon 64 FX series. Itís a small nod to enthusiasts, even if the architecture isnít able to overclock well.
All of the other Smithfield-specific features persist. The Extreme Edition chip has 1MB of L2 cache per core, for 2MB total, it operates on an 800 MHz front side bus, it supports Execute Disable, and the chip features EM64T technology--a real looker now that Microsoft has put the wraps on its 64-bit version of Windows XP. Enhanced Intel SpeedStep Technology, featured on the Pentium D processors, is not listed as one of the Extreme Editionís attributes.
Manufactured at 90nm, itís little surprise that the Extreme Edition 840 is fairly gargantuan at 206 square millimeters. It consists of 230 million transistors, and if you look at a die plot of the processor, you can clearly see two distinct cores. Such a massive die is undoubtedly very expensive to manufacture, especially considering the yield consideration involved with using two cores on one die. Needless to say, the move to 65nm canít come soon enough for Intel.
Finally, because weíre dealing with a preview here, Intel is choosing to keep pricing information secret. Gut instinct, along with informal conversations, suggest that the chip will cost $1,000. Though formerly an exorbitant price, Intelís Extreme Edition 840 would be an attractive option for a small business server or workstation looking to go multi-processor without spending a load of money. Believe it or not, such a platform would actually be less expensive than a traditional dual-processor configuration.
| Intel 955X Express||Page:: ( 5 / 14 )|
A New Platform Already?
Only six months have passed since Intel released its last flagship chipset, the 925XE (will we ever see another chipset with the longevity of Intelís 440BX?).Weíd be elated if the Pentium D dropped right into one of those 925-based motherboards--theyíre both LGA775, after all--but the processor indeed requires an entirely new platform with special concessions for dual-core processors.
|<% print_image("05"); %>||<% print_image("06"); %>|
Fortunately, the 955X Express chipset includes more than just support for dual-core processors. The MCH now boasts official DDR2-667 memory compliance, capacities up to 8GB, and ECC error correction. Our D955XBK motherboard features four, 64-bit slots, suggesting a need for 2GB modules in order to top the boardís ceiling. Flex Memory Technology makes it possible to install memory modules of different sizes, preserving dual-channel mode even when youíre using mismatched sizes.
Intelís documentation claims 955X only supports 800 and 1066 MHz front side bus speeds. The D955XBKís BIOS also lists 533MHz, though, so perhaps LGA775 Celeron processors would work as well.
|<% print_image("07"); %>||<% print_image("08"); %>||<% print_image("09"); %>|
Another interesting feature is a second x16 PCI Express graphics port on the reference motherboard. Although the 955X MCH only offers 16 lanes of connectivity, Intelís new ICH7 I/O hub enables six x1 lanes, four of which are aggregated together to make up that slot mechanically plumbed for a video card. We tried installing a pair of GeForce 6600 GT cards and linking them with an SLI connector to no avail. Even without the connector, Intelís D955XBK refused to recognize the second card.
The only other notable hardware addition is support for Serial ATA 3 Gbps across the ICH7ís four ports. Of course, NCQ, introduced with the ICH6, is still a standard feature. Intelís Matrix Storage Technology now allows RAID 0, 1, 5, and 10, establishing a cost-effective alternative to expensive SATA RAID add-in cards.
|<% print_image("10"); %>||<% print_image("11"); %>||<% print_image("12"); %>|
Power, power, power
One Prescott core, dissipating in excess of 110W of power, is already a challenge to keep cool. Adding a second core takes even more effort. At 3.2 GHz, the Pentium Extreme Edition 840 has a thermal design power of 130W. No longer is the standard ATX power connector and 12V auxiliary input sufficient, it seems. According to Intelís setup documentation, you either need a 20-pin ATX output, the 12V connector and another four-pin power connector or a 24-pin ATX plug and the four-pin Molex plug. Our sample system came with the 24-pin ATX connector and an eight-pin EPS output, the kind youíd normally find on a dual-processor server or workstation.
|<% print_image("13"); %>||<% print_image("14"); %>|
|<% print_image("15"); %>||<% print_image("16"); %>|
| System Setup||Page:: ( 6 / 14 )|
Intel Pentium Extreme Edition 840 (3.2 GHz)
Intel Pentium 4 570J (3.8 GHz)
Intel Pentium 4 540J (3.2 GHz)
AMD Athlon 64 FX-55 (2.6 GHz)
AMD Athlon 64 4000+ (2.4 GHz)
Intel D955XBK 955X Express Motherboard
Intel D925XECV2 925XE Express Motherboard
ASUS A8N-SLI nForce4 SLI Motherboard
1GB Micron DDR2-667 CAS 5 (2x512MB)
1GB Corsair DDR2-533 CAS 3 Pro Series Memory (2x512MB)
1GB Corsair DDR-400 CAS 2 Xpert Series Memory (2x512MB)
ATI RADEON X850 XT
Seagate Barracuda 7200.8 400GB
Windows XP Professional with Service Pack 2
Desktop resolution 1024x768, 32-bit color, 85Hz refresh
All power saving options were turned off, as were the Automatic Update and System Restore services. Graphics options under the ĎPerformanceí tab were all disabled for maximum performance.
As youíll see, weíre again doing a few things differently. Baseline benchmarks now begin at 1024x768. Although itís certainly relevant to show off performance at 800x600 or 640x480, where youíd typically see the greatest differences between processor and platforms, those settings are really falling by the wayside and in no way reflect actual game play. Youíll be surprised nonetheless. Even at 1024x768, thereís plenty of difference between competing chips.
Windows Media Encoder 9
SiSoft Sandra 2005
| Windows Media Encoder 9||Page:: ( 7 / 14 )|
Microsoft Windows Media Encoder 9
Media encoding is one of those usages that Intel claims will really showcase the capability of dual-core processing. In fact, the Smithfield demo at CES was running a copy of Windows XP Media Center Edition 2005 with playback and encoding tasks running simultaneously (in HD, no less!). Testing just the encoding with WME 9, itís very clear that performance stands to improve greatly thanks to Intelís dual-core solution, as the 3.2 GHz Extreme Edition 840 handily bests Intelís 3.8GHz 570J and everything else in our comparison.
| 3DMark05||Page:: ( 8 / 14 )|
The gaming component of 3DMark isnít optimized for threaded operation, which is why you see Intelís dual-core 3.2 GHz chip scoring similarly to the older 3.2 GHz Pentium 4. AMD maintains its grip on gaming performance, followed by the 3.8 GHz Pentium 4 delivering better single-threaded results.
Itís a different story entirely when the processor test runs. Intelís Pentium Extreme Edition 840 takes a commanding lead, demonstrating superiority over the 3.8 GHz 570J, even though the Prescott chip supports threaded operation through Hyper-Threading. AMDís Athlon 64 FX-55 uses brute force to course past the single-core 3.2 GHz Pentium 4 540.
| PCMark04||Page:: ( 9 / 14 )|
We chose to run three different components of Futuremarkís PCMark04 test suite, including the system, processor, and memory sections. Intel says that the processor suite is the only one with threaded content; however, the dual-core chip fares fairly well in the overall system metric too, succumbing to Intelís 3.8 GHz chip just barely.
The tables turn during our CPU test runs, where the dual-core Extreme Edition hops out in front, overshadowing Intelís other offerings, in addition to the Athlon 64 FX.
The memory test doesnít turn out quite so positively. Even though weíre using DDR2-667, per Intelís recommendation, high latency seems to really be hurting overall throughput. The 3.8 GHz chip turns in our highest scores thanks to CAS 3 timings and 533 MHz frequencies.
| Cinebench 2003||Page:: ( 10 / 14 )|
Maxon Cinema 4D Cinebench 2003
Rendering software is right up there with media encoding on Intelís roster of threaded applications geared to pick up plenty of performance thanks to dual-core. Indeed, Cinebench 2003 successfully identifies the two physical cores and four logical cores, splitting its workload between them. Consequentially, the Extreme Edition 840 simply flies through this test, destroying the other processors.
| Doom 3||Page:: ( 11 / 14 )|
id Software Doom 3
Intelís own documents list Doom 3 as a single-threaded app incapable of exploiting dual-core benefits. Nevertheless, at high resolutions, processor performance isnít holding us back, anyway. Even if Doom 3 were optimized for multiple threads, the X850 XT in our test bed is the bottleneck.
Down low, at 1024x768, where processor performance makes more of a difference, the Extreme Edition 840 performs closer to the Pentium 4 540, while AMDís architecture reigns supreme. Resolution is ultimately the equalizer here, though.
| Half-Life 2||Page:: ( 12 / 14 )|
Valve Half-Life 2
Thereís little secret which processor architecture is favored in Half-Life 2.
The dual-core chip actually drops to last place here, in both high- and low-resolution tests. Even Intelís older 3.2 GHz chip picks up extra speed.
| SiSoft Sandra 2005||Page:: ( 13 / 14 )|
SiSoft Sandra 2005
| Conclusion||Page:: ( 14 / 14 )|
There are already many environments where a dual-core processor will outclass a non-threaded chip, such as media encoding, rendering, and the everyday multi-tasking in which you engage, behavior thatís very hard to quantify. Intel tells the tale of a ďperfect stormĒ--instances during your work day where several tasks execute simultaneously, both in the background and foreground, taxing your system. Although you might be idling 90 percent of the time, those perfect storms can be debilitating. Without a processor capable of addressing multiple threads simultaneously, youíre left waiting. And thatís where dual-core comes into play.
Of course, itís hard to believe some of the hypothetical scenarios that Intel puts forth. Youíd probably never play Need for Speed while recording two TV shows simultaneously, for instance. But once game developers get their hands on this technology and start writing software to employ it, more immersive environments should be the end result. Weíll see. With rumblings of late 2006, threaded games are undoubtedly still along way off.
Over the course of the next several weeks, weíll be refining our testing methodologies, looking for applications and multitasking procedures that exploit dual-core technology, as well as those that do not. Weíll be reporting our findings in follow-up articles, including overclocking. The ramifications of having multiple cores on one processor die are unknown, but as anyone thatís attempted overclocking a dual processor system can tell you, overclocking can be a challenge.
Intel hasnít provided pricing information for the Pentium Extreme Edition 840 or the Pentium D line, but Intel says the first chips should begin shipping later this quarter; todayís ďlaunchĒ is meant to be a sneak peek at the technology. Already we can see some cases where dual-core really shines, but is the technology ready for primetime? Only time will tellÖ